A persistent effort in neuroscience has been to pinpoint the neurobiological substrates that support mental processes. The Research Domain Criteria (RDoC) aims to develop a new framework based on fundamental neurobiological dimensions. However, results from several meta-analysis of task-based fMRI showed substantial spatial overlap between several mental processes including emotion and anticipatory processes, irrespectively of the valence. Consequently, there is a crucial need to better characterize the core neurobiological processes using a data-driven techniques, given that these analytic approaches can capture the core neurobiological processes across neuroimaging literature that may not be identifiable through expert-driven categories. Therefore, we sought to examine the main data-driven co-activation networks across the past 20 years of published meta-analyses on task-based fMRI studies. We manually extracted 19,822 coordinates from 1,347 identified meta-analytic experiments. A Correlation-Matrix-Based Hierarchical Clustering was conducted on spatial similarity between these meta-analytic experiments, to identify the main co-activation networks. Activation likelihood estimation was then used to identify spatially convergent brain regions across experiments in each network. Across 1,347 meta-analyses, we found 13 co-activation networks which were further characterized by various psychological terms and distinct association with receptor density maps and intrinsic functional connectivity networks. At a fMRI activation resolution, neurobiological processes seem more similar than different across various mental functions. We discussed the potential limitation of linking brain activation to psychological labels and investigated potential avenues to tackle this long-lasting research question.